chroma-core / docs Goto Github PK

Ref chroma-core/chroma#1512

Document best practices for comparing floats using where filters. I.E Don't use $eq and use $gt and $lt instead with a tolerance.

If a key is not included in a record, but is in the where filter, the record will be ignored by the where filter, there is not implicit value.

https://docs.trychroma.com/api-reference#in-memory-chroma-with-savingloading-to-disk

currently they are not very good!

consider https://python.langchain.com/docs/integrations/vectorstores/chroma is much better

The current docs use JS with positional params, we are switching over to named.

Update the docs to reflect that change.

This will be merged in Monday the 22nd alongside the release.

eg https://github.com/mariochavez/chroma

When first trying Chroma, I ran into an issue with Python 3.12.x on an ARM M2 Mac and had to downgrade to 3.11.5. Other people have had the same issue [1]. If required, users should be told upfront which is the recommended version.

Request: Add Docs to specify the current working Python version.

[1] Related Issue:
chroma-core/chroma#1410 chroma-core/chroma#163

https://discord.com/channels/1073293645303795742/1172062624733728768

Fixing the documentation issue as mentioned in chroma-core/chroma#1062

Fix heading in python API docs so the sidebar works.

I think we need to add this to pythonDocs.sh and just manually look for #### and replace with ###

self explanatory

sentence_transformer_ef = embedding_functions.SentenceTransformerEmbeddingFunction(
model_name="intfloat/multilingual-e5-large")

be more specific that you can use custom models with this embedding fn

https://discord.com/channels/1073293645303795742/1133432402425938031/1136931539368825013

Help users understand the perf and memory req of the system at various scales

We chose to defer docs for

chroma-core/chroma#1244

Because the api-doc-gen doesn't work well with the inheritance changes on ClientAPI/ServerAPI/BaseAPI. We should add these doc changes

chroma-core/chroma#221 (comment)

We should add this tip to our documentation.

right now they are all very different. maybe we include info like

• description
• dimensionality
• max input tokens
• tokenizer
• cost
• languages (english, other?)
• performance stats? on MTEB/BEIR?
• decoder
• how to get an API Key
• link to their docs / marketing

The User Guide states:

Each document must have a unique associated id. If you try to .add a document with an id that already exists in the collection, it will raise an exception.

The below example demonstrates that this is not the case:

import chromadb

client = chromadb.Client()
collection = client.create_collection(name="my_collection")
collection.add(
    documents=['doc1', 'doc2'],
    ids=['id1', 'id1'] # <-- duplicate IDs
)

collection.get(
    ids=["id1"]
)

Result:

{'ids': ['id1', 'id1'],
 'embeddings': None,
 'documents': ['doc1', 'doc2'],
 'metadatas': [None, None]}

Inspecting the implementation for inserting new records in the add() implementation for duckdb and the add() implementation for clickhouse appear to support the statement that uniqueness is not enforced and no exception will be raised.

I’m new to this project and wanted to create Embeddings using val = default_ef("foo") and was getting absolutely massive embedding sizes. I think the string “foo” gets turned into a list of its chars and then each char gets an embedding. I think it should be val = default_ef(["foo"]) instead.